Skip to Main Content
The stereo-matching algorithm based on adaptive-weight can produce disparity map of high precision. However it is not suitable for real-time application due to high computational complexity. Wang etc apply one pass aggregation to adaptive-weight algorithm and can achieve real-time performance but with a much lower precision. This paper presents a new real-time adaptive weight stereo matching algorithm based on two-pass aggregation: Matching costs are aggregated by row and array in order to reduce the computational complexity, some measures are taken to reduce the accuracy lost; then a fast dynamic programming combined greedy strategy and disparity smoothness constraint is adopted to select disparity. Evaluation results in MiddleBury platform show that the algorithm ranked No.2 among all real-time algorithms in this platform and the average error rate is the smallest. Also this algorithm achieves a good real-time performance.